A point cloud data segmentation method and system based on a DBSCAN clustering algorithm

Abstract

The application provides a point cloud data segmentation method and system based on a DBSCAN clustering algorithm, and belongs to the technical field of data processing.The method comprises the following steps: obtaining non-ground point cloud data to obtain final high-density data and final low-density data; obtaining DBSCAN clustering algorithms under a plurality of parameter combinations; obtaining a first reference value and a second reference value in the fitness of a clustering result under each parameter combination according to a clustering cluster, a core point and a noise point under each parameter combination, and then obtaining an initial clustering cluster and an initial noise point; and clustering the initial noise point to obtain a final clustering result.The application adopts a secondary clustering and distance checking mode, avoids misjudgment of low-density objects as noise, and reduces missed segmentation; through a processing mode of first partitioning, then targeted clustering, and finally checking and optimizing, the influence of uneven point cloud density on DBSCAN clustering is effectively overcome, and the accuracy of non-ground point cloud segmentation is improved.

Description

Technical Field

[0001] This invention belongs to the field of data processing, specifically relating to a point cloud data segmentation method and system based on the DBSCAN clustering algorithm. Background Technology

[0002] Point cloud data is a collection of discrete 3D spatial points acquired through 3D scanning equipment. Each point typically contains coordinate information and may also include attributes such as color, intensity, and normal vectors. It is widely used in computer vision, robotics, and autonomous driving. Because road surface point cloud data contains complex scene mixing information, it is necessary to segment the data. Currently, when segmenting point cloud data using the RANSAC algorithm, it only divides the data into ground point cloud data and non-ground point cloud data, without further classifying the non-ground point cloud data. Therefore, clustering is performed on the non-ground point cloud data.

[0003] Clustering analysis is a fundamental technique in data mining, aiming to divide objects in a dataset into multiple groups based on similarity, achieving high similarity within groups and significant differences between groups. Among these, the density-based clustering algorithm DBSCAN dynamically determines the number of clusters based on changes in dataset density, eliminating the need for pre-specifying the number of clusters and recognizing clusters of arbitrary shapes and sizes. Therefore, DBSCAN is used for further classification of non-ground point cloud data. However, when acquiring road surface point cloud data, the closer an object is to the 3D point cloud data acquisition device, the more 3D point cloud data of that object is collected, resulting in point cloud data containing various densities. This makes it impossible to use a single parameter combination of neighborhood radius and minimum number of points to obtain the optimal clustering result in a single clustering operation when using DBSCAN to cluster point cloud data. Furthermore, DBSCAN requires manual setting of the neighborhood radius and minimum number of points, leading to significant discrepancies between the segmentation results and the actual situation when using DBSCAN to segment non-ground point cloud data. Summary of the Invention

[0004] To address the problem that the clustering results of the DBSCAN clustering algorithm for classifying non-ground point cloud data differ significantly from the classification of real-world non-ground point cloud objects, this invention provides a point cloud data segmentation method based on the DBSCAN clustering algorithm.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] Acquire non-terrestrial point cloud data;

[0007] Based on the distribution of non-ground point cloud data, the non-ground point cloud data is divided into final high-density data and final low-density data.

[0008] While ensuring that the core points are the final high-density data, the DBSCAN clustering algorithm is used to cluster all non-ground point cloud data to obtain initial clusters and initial noise points.

[0009] While ensuring that the core points are the final low-density data, the DBSCAN clustering algorithm is used to cluster the initial noise points to obtain noise clusters and secondary noise points;

[0010] The final clustering result is obtained based on the Euclidean distance between the secondary noise points and the initial cluster, as well as the distance between each data point within the noise cluster, thus completing the segmentation of the non-terrestrial point cloud data.

[0011] This invention, based on point cloud distribution characteristics, divides non-terrestrial point clouds into "final high-density data" (near-range dense points, such as nearby objects) and "final low-density data" (far-range sparse points, such as distant objects) according to density thresholds. This breaks the limitation of "processing full-density data with a single parameter," providing a basis for subsequent parameter-based clustering and serving as a prerequisite for solving density unevenness. For the "final high-density data," parameters suitable for dense points are applied to cluster into initial clusters (mainly corresponding to nearby objects), while initial noise points (which may include low-density points or true noise) are labeled. Focusing on high-density regions, clear nearby objects are captured first, avoiding excessive merging of dense points due to overly broad parameters. For the "initial noise points" from the first clustering, the "final low-density data" is selected, and secondary clustering is performed using parameters suitable for sparse points to obtain noise clusters (corresponding to distant sparse objects). For low-density regions missed in the first clustering, distant objects are identified through parameter adaptation, avoiding omissions due to overly strict parameters. Furthermore, the Euclidean distance between secondary noise points (points not clustered in the second cluster) and the initial cluster and the noise cluster is calculated. This distance verification compensates for the boundary errors of the two clustering processes, ensuring the correct classification of edge points and sparse points.

[0012] Preferably, the step of dividing the non-ground point cloud data into final high-density data and final low-density data based on the distribution of the non-ground point cloud data involves the following steps:

[0013] Calculate the first The Euclidean distance between the nth non-ground point cloud data and all non-ground point cloud data is obtained. A distance set of non-terrestrial point cloud data;

[0014] The first Let denoted as the set of all data corresponding to the minimum distance within the distance set of the nth non-ground point cloud data. The nearest neighbor data of a non-terrestrial point cloud data;

[0015] Through the first The non-ground point cloud data and its first Given a set of nearest neighbor data, obtain a nearest neighbor pair.

[0016] The Euclidean distance between the two data points in each nearest neighbor pair is recorded as the nearest neighbor distance of that nearest neighbor pair, thus obtaining the nearest neighbor distance set;

[0017] Each nearest neighbor distance in the nearest neighbor distance set is used as a segmentation threshold, and high-density data and low-density data under each segmentation threshold are obtained based on the segmentation threshold.

[0018] Based on the high-density and low-density data within each nearest neighbor pair under each segmentation threshold, the final high-density data and the final low-density data are obtained.

[0019] Preferably, the specific steps for obtaining the high-density data and low-density data under each segmentation threshold are as follows:

[0020] Let the segmentation threshold be the first... The nearest neighbor distance is defined as the non-ground point cloud data pair within the nearest neighbor data pair whose nearest neighbor distance is greater than the segmentation threshold.

[0021] Non-ground point cloud data pairs within the nearest neighbor data pairs whose nearest neighbor distance is less than or equal to the segmentation threshold are denoted as high-density data pairs.

[0022] Preferably, the specific steps for obtaining the final high-density data and the final low-density data based on the high-density data and low-density data within each nearest neighbor data pair under each segmentation threshold are as follows:

[0023] Obtain the segmentation threshold as the first The segmentation effectiveness parameter for the nearest neighbor distance is calculated using the following formula:

[0024] ;

[0025] In the formula, The segmentation threshold is the first... The segmentation effectiveness parameter when the nearest neighbor distance is given. This indicates the number of data points belonging to high-density data pairs. This indicates the number of data points belonging to low-density data pairs. For the first The number of nearest neighbor data point pairs whose nearest neighbor distance is less than the segmentation threshold within all nearest neighbor data pairs of a data point belonging to a high-density data pair; For the first The number of nearest neighbor pairs whose nearest neighbor distance is greater than the segmentation threshold within all nearest neighbor pairs of data points belonging to low-density data pairs;

[0026] The nearest neighbor distance corresponding to the maximum segmentation effectiveness is denoted as the optimal segmentation threshold;

[0027] Non-ground point cloud data within the nearest neighbor data pair whose nearest neighbor distance is greater than the optimal segmentation threshold are denoted as final low-density data; non-ground point cloud data within the nearest neighbor data pair whose nearest neighbor distance is less than or equal to the optimal segmentation threshold are denoted as final high-density data.

[0028] Preferably, the method further includes combining the non-dominated sorting genetic algorithm with the DBSCAN clustering algorithm to obtain the optimal parameter combination when using the DBSCAN clustering algorithm to cluster non-ground point cloud data, and obtaining the final segmentation result by using the optimal parameter combination when using the DBSCAN clustering algorithm to cluster non-ground point cloud data.

[0029] Preferably, the optimal parameter combination used in the DBSCAN clustering algorithm to cluster non-ground point cloud data to obtain the final segmentation result includes the following specific steps:

[0030] Using a non-dominated sorting genetic algorithm, multiple parameter combinations are obtained. Each parameter combination contains a high-density data parameter combination with a neighborhood radius and a minimum number of points, and a low-density parameter combination with a neighborhood radius and a minimum number of points. The neighborhood radius and minimum number of points in the high-density parameter combination are different from those in the low-density parameter combination.

[0031] The neighborhood radius and minimum number of points under the high-density data parameter combination within the y-th parameter combination are taken as the neighborhood radius and minimum number of points within the DBSCAN clustering algorithm, thus obtaining the DBSCAN clustering algorithm under the high-density parameter combination within the y-th parameter combination.

[0032] The neighborhood radius and minimum number of points under the low-density data parameter combination within the y-th parameter combination are taken as the neighborhood radius and minimum number of points within the DBSCAN clustering algorithm, thus obtaining the DBSCAN clustering algorithm under the low-density parameter combination within the y-th parameter combination.

[0033] While ensuring that the core points are the final high-density data, the DBSCAN clustering algorithm under the high-density parameter combination within the y-th parameter combination is used as the distance metric to cluster all non-ground point cloud data, and the clusters, core points and noise points under the high-density parameter combination within the y-th parameter combination are obtained. These are denoted as the high-density cluster within the y-th parameter combination, and the noise points and core points within this parameter combination.

[0034] While ensuring that the core points are the final low-density data, the DBSCAN clustering algorithm under the low-density parameter combination in the y-th parameter combination is used to continue clustering the noise points in the y-th parameter combination. The clusters, noise points and core points obtained from this clustering are respectively denoted as the low-density clusters in the y-th parameter combination, the secondary noise points in the y-th parameter combination and the core points in the y-th parameter combination.

[0035] Combine the y-th parameter into the th... The Euclidean distance between each secondary noise point and each data point within the initial cluster and the noise cluster is denoted as the distance between the y-th parameter combination and the y-th data point. The difference distance between each secondary noise point;

[0036] If the y-th parameter combination contains the th... If the minimum difference distance of the _th secondary noise points is less than or equal to the neighborhood radius of the high-density data parameter combination within the _th parameter combination, then the _th parameter combination within the _th parameter combination will be... Each secondary noise point is placed within the cluster of data whose minimum difference distance corresponds to its own.

[0037] Perform the above operation on all secondary noise points in the y-th parameter combination to obtain the updated low-density cluster and high-density cluster in the y-th parameter combination, and record them as the final cluster in the y-th parameter combination.

[0038] By analyzing the distribution of data around the core points within the final cluster within the y-th parameter combination, we obtain the first reference value of the final clustering result within the y-th parameter combination.

[0039] Based on the distance between the core points in different final clusters within the y-th parameter combination, the distribution differences of the data around the core points, and the distribution of noise points, a second reference value for the final clustering result within the y-th parameter combination is obtained.

[0040] The first and second reference values ​​of the final clustering result in each parameter combination are used as the two fitness values ​​of that parameter combination and input into the non-dominated sorting genetic algorithm to obtain the optimal parameter combination when using the DBSCAN clustering algorithm to cluster non-ground point cloud data.

[0041] When using the DBSCAN clustering algorithm to cluster non-ground point cloud data, the neighborhood radius and minimum number of points in the optimal parameter combination are used as the neighborhood radius and minimum number of points in the DBSCAN clustering algorithm, thus obtaining the DBSCAN clustering algorithm under the high-density parameter combination within the optimal parameter combination.

[0042] While ensuring that the core points are the final high-density data, the DBSCAN clustering algorithm under the high-density parameter combination within the optimal parameter combination is used as the distance metric to cluster all non-ground point cloud data, and the clusters and noise points under the high-density parameter combination within the optimal parameter combination are obtained, which are denoted as the initial clusters and initial noise points, respectively.

[0043] The neighborhood radius and minimum number of points under the low-density data parameter combination within the optimal parameter combination when clustering all non-ground point cloud data are taken as the neighborhood radius and minimum number of points within the DBSCAN clustering algorithm, thus obtaining the DBSCAN clustering algorithm under the low-density parameter combination within the optimal parameter combination.

[0044] While ensuring that the core points are the final low-density data, the DBSCAN clustering algorithm under the low-density parameter combination within the optimal parameter combination is used as the distance metric to cluster all non-ground point cloud data, and the clusters and noise points under the low-density parameter combination within the optimal parameter combination are obtained, which are denoted as noise clusters and secondary noise points, respectively.

[0045] The final clustering result is obtained based on the Euclidean distance between the secondary noise points and the initial cluster, as well as the distance between each data point within the noise cluster.

[0046] Preferably, the first reference value of the final clustering result within the y-th parameter combination is obtained by analyzing the distribution of data around the core point within the final cluster within the y-th parameter combination. The specific steps are as follows:

[0047] The combination with the y-th parameter is the first Within the final cluster, the first The data whose Euclidean distance to each core point is less than or equal to the neighborhood radius under the high-density parameter combination within the y-th parameter combination is denoted as the data within the y-th parameter combination. Within the final cluster, the first Neighborhood points of a core point;

[0048] Combine the y-th parameter into the th... Within the final cluster, the first The mean of the Euclidean distances between each core point and all its neighboring points is used as the y-th parameter combination. Within the final cluster, the first Density estimates for each core point;

[0049] Combine the y-th parameter into the th... The variance of the density estimates of all core points within the final cluster is denoted as the variance of the nth parameter combination within the yth parameter combination. The density consistency of each final cluster;

[0050] The sum of the density consistency of all final clusters within the y-th parameter combination is denoted as the first reference value of the final clustering result within the y-th parameter combination.

[0051] Preferably, the second reference value for the final clustering result within the y-th parameter combination is obtained based on the distance between the core points in different final clusters within the y-th parameter combination, the distribution differences of the data around the core points, and the distribution of noise points. The specific steps are as follows:

[0052] Calculate the y-th parameter combination. Each core point within the final cluster is related to the first... The distance between each core point within each final cluster;

[0053] If the y-th parameter combination contains the th... Within the final cluster, the first Within the combination of the _ core point and the y-th parameter, the _ _ Within the final cluster, the first If the Euclidean distance between two core points is less than 2 times the neighborhood radius under the high-density parameter combination within the y-th parameter combination, then the two clusters are considered adjacent. Then, the difference between the Euclidean distance between the core points of these two different clusters and the neighborhood radius under the high-density parameter combination under the y-th parameter combination is used as a parameter for over-subdivision scoring.

[0054] Get the y-th parameter combination. The final cluster and the first The specific formula for calculating the subdivision score of the final cluster is as follows:

[0055] ;

[0056] In the formula, Within the y-th parameter combination, the th... The final cluster and the first The subdivision score of the final cluster, Indicates the y-th parameter combination. The variance of the density estimates of the core points within each final cluster Indicates the y-th parameter combination. The variance of the density estimate of the core within each final cluster. Indicates the y-th parameter combination. The final cluster and the first The number of subdivisions in the final cluster. This represents the neighborhood radius of the high-density data parameter combination within the y-th parameter combination. Indicates the y-th parameter combination. The final cluster and the first The first final cluster Each level of detailed performance Represents the absolute value function;

[0057] The sum of the over-subdivision scores of all two different final clusters within the y-th parameter combination is input into the normalization function, and the output is recorded as the over-subdivision score of the final clustering result within the y-th parameter combination.

[0058] The formula for calculating the noise ratio of the final clustering result within the y-th parameter combination is as follows:

[0059] ;

[0060] In the formula, This represents the noise proportion of the final clustering result within the y-th parameter combination. This indicates the amount of non-terrestrial point cloud data. This represents the number of non-ground point cloud data points that are not in the final cluster within the y-th parameter combination;

[0061] The specific calculation formula for obtaining the second reference value within the fitness of the clustering results under the high-density parameter combination within the y-th parameter combination is as follows:

[0062] ;

[0063] In the formula, The second parameter represents the final clustering result within the y-th parameter combination. This represents the over-subdivision score of the final clustering result within the y-th parameter combination. This represents the noise proportion of the final clustering result within the y-th parameter combination.

[0064] Preferably, the final clustering result is obtained based on the Euclidean distance between the secondary noise points and the initial cluster, as well as the distance between each data point within the noise cluster. The specific steps are as follows:

[0065] The first The Euclidean distance between the nth secondary noise point and each data point within the initial cluster and the noise cluster is denoted as the nth noise point. The difference distance between each secondary noise point;

[0066] If the first If the minimum difference distance of the nth secondary noise points is less than or equal to the neighborhood radius of the DBSCAN clustering algorithm under the optimal high-density parameter combination when clustering all non-ground point cloud data, then the nth... Each secondary noise point is placed within the cluster of data whose minimum difference distance corresponds to its own.

[0067] Perform the above operation on all secondary noise points to obtain the updated initial cluster and the updated noise cluster; based on the updated initial cluster and the noise cluster, obtain the final clustering result.

[0068] This invention also proposes a point cloud data segmentation system based on the DBSCAN clustering algorithm, comprising:

[0069] Point cloud data acquisition module, used to acquire non-ground point cloud data;

[0070] The point cloud data segmentation module is used to divide the non-ground point cloud data into final high-density data and final low-density data according to the distribution of the non-ground point cloud data.

[0071] While ensuring that the core points are the final high-density data, the DBSCAN clustering algorithm is used to cluster all non-ground point cloud data to obtain initial clusters and initial noise points.

[0072] While ensuring that the core points are the final low-density data, the DBSCAN clustering algorithm is used to cluster the initial noise points to obtain noise clusters and secondary noise points;

[0073] The final clustering result is obtained based on the Euclidean distance between the secondary noise points and the initial cluster, as well as the distance between each data point within the noise cluster, thus completing the segmentation of the non-terrestrial point cloud data.

[0074] The point cloud data segmentation method based on the DBSCAN clustering algorithm provided by this invention has the following beneficial effects: When clustering non-ground point cloud data, this invention, based on the characteristic that the amount of point cloud data of an object is inversely proportional to the distance to the LiDAR when acquiring point cloud data of an object via LiDAR, first clusters the non-ground point cloud data when obtaining the final clustering result, and then clusters the noise points within the clustering result. This solves the problem that, since the amount of point cloud data of an object is inversely proportional to the distance to the LiDAR, i.e., the closer an object is to the LiDAR, the more LiDAR point cloud images of that object are, the more likely the object will be to be identified as noise when directly clustering using the DBSCAN clustering algorithm. Based on the characteristic that the DBSCAN clustering algorithm essentially clusters data points based on the distribution of other data points around each data point, this invention uses the distribution of data points around the core point within each cluster as a parameter to evaluate the effectiveness of the clustering result, thus solving the problem of current evaluation methods for clustering results. The current method of evaluating clustering results is not suitable for assessing the effectiveness of clustering results obtained by the DBSCAN algorithm. Evaluating clustering quality by the distribution of data points around the core points within each cluster may lead to over-clustering. A second parameter for evaluating the effectiveness of clustering results is obtained by analyzing the distance distribution of core points within different clusters, thus avoiding over-clustering. Subsequently, using the two parameters for evaluating the effectiveness of clustering results when clustering non-terrestrial point cloud data under each parameter combination, a non-dominated sorting genetic algorithm is used to obtain the optimal clustering result for the initial clustering. Then, a second clustering is performed on the noisy points within the initial clustering result. This addresses the problem that when acquiring point cloud data via LiDAR, the amount of point cloud data is inversely proportional to the distance between the LiDAR and the object, causing the current method's clustering results to differ significantly from reality. It also addresses the issue that existing evaluation criteria for clustering results cannot adequately assess the clustering results obtained by the DBSCAN algorithm, leading to unsatisfactory segmentation results when using the DBSCAN clustering method for point cloud data segmentation. Attached Figure Description

[0075] To more clearly illustrate the embodiments and design schemes of the present invention, the accompanying drawings required for this embodiment will be briefly described below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0076] Figure 1 This is a flowchart of the point cloud data segmentation method based on the DBSCAN clustering algorithm in Embodiment 1 of the present invention;

[0077] Figure 2 The target area is a real-world scene image;

[0078] Figure 3 Point cloud image of the target region;

[0079] Figure 4 This is the initial segmentation image of the point cloud image of the target region;

[0080] Figure 5 This is a schematic diagram illustrating the results of clustering data using the DBSCAN clustering algorithm; where, Figure 5 (a) shows the clustering results when the data is clustered using the DBSCAN clustering algorithm. Figure 5 (b) shows the clustering results when the data is clustered using a traditional clustering algorithm;

[0081] Figure 6 This is a schematic diagram illustrating the results of over-clustering;

[0082] Figure 7 This is a schematic diagram showing the distribution of noise points and data points within clusters after clustering; where, Figure 7 Image (a) shows the clustering results of the initial clustering of noise points. Figure 7 (b) is for Figure 7 The clustering result image after redistributing the noise points in Figure (a);

[0083] Figure 8 This is a diagram showing the results of clustering non-terrestrial point cloud data using existing methods.

[0084] Figure 9 This is a diagram showing the result of clustering non-terrestrial point cloud data using the method of the present invention.

[0085] Figure 10 This is a diagram showing the results of dividing high-density and low-density data. Detailed Implementation

[0086] To enable those skilled in the art to better understand and implement the technical solutions of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.

[0087] Example 1

[0088] This invention provides a point cloud data segmentation method based on the DBSCAN clustering algorithm, specifically as follows: Figure 1 As shown, it includes:

[0089] Step S001: Acquire non-ground point cloud data.

[0090] Specifically, lidar is used to acquire lidar point cloud images of the target area on the road surface, and the RANSAC algorithm is used to divide the lidar point cloud images of the target area into non-ground point cloud data and ground point cloud data. Figure 2 The target area is a real-world scene image. Figure 3 For the point cloud image of the target region, Figure 4 This is the initial segmentation image of the point cloud image of the target region. Figure 2 The object contained within the left square in the middle and Figure 3 The objects within the left-hand box all belong to the same car. Figure 2 The objects contained within the second box on the left and Figure 3 The objects contained within the two boxes on the left in the image are both the same person. Figure 4 In the diagram, red pixels represent non-ground point cloud data, blue pixels represent ground point cloud data, and white areas represent blind spots for the LiDAR, where the LiDAR does not obtain point cloud data. The use of the RANSAC algorithm to separate point cloud data into ground and non-ground point cloud data is a well-known existing technique and will not be elaborated upon in this embodiment.

[0091] Step S002: Based on the distribution of non-ground point cloud data, obtain several nearest neighbor data pairs and segmentation thresholds, and then obtain high-density data and low-density data under each segmentation threshold; based on the high-density data and low-density data in each nearest neighbor data pair under each segmentation threshold, divide the non-ground point cloud data into final high-density data and final low-density data.

[0092] It's important to note that when acquiring 3D point cloud data using LiDAR, the density of the point cloud data for an object is related to the distance between the object and the LiDAR. That is, when collecting 3D point cloud data with LiDAR, the collected point cloud data exhibits a pattern of denser data at closer distances and sparser data at farther distances. Therefore, when using the DBSCAN clustering algorithm to cluster non-terrestrial point cloud data, it's necessary to first divide the non-terrestrial point cloud data into high-density and low-density data before applying the DBSCAN clustering algorithm to perform the clustering.

[0093] It should be further explained that when dividing non-ground point cloud data into high-density and low-density data, ideally, high-density data should be surrounded by other high-density data, and low-density data should be surrounded by other low-density data. Therefore, multiple segmentation thresholds are used to divide non-ground point cloud data into high-density and low-density data. After each segmentation, the segmentation result at each threshold is determined by checking if each data point is of the same type as its nearest neighbor. The optimal threshold is then obtained, leading to the segmentation result under the optimal threshold.

[0094] Specifically, calculate the first The distance between the nth non-ground point cloud data and all non-ground point cloud data is obtained. The distance set of the non-ground point cloud data will be the first Let denoted as the set of all data corresponding to the minimum distance within the distance set of the nth non-ground point cloud data. The nearest neighbor data of a non-terrestrial point cloud dataset. Through the first... The non-ground point cloud data and its first Given a set of nearest neighbor data points, we obtain a nearest neighbor pair. The Euclidean distance between the two data points in each nearest neighbor pair is denoted as the nearest neighbor distance for that pair. This results in a set of nearest neighbor distances. The nearest neighbor distance set does not contain nearest neighbors with the same numerical value.

[0095] At this point, we have obtained all the nearest neighbor pairs and the nearest neighbor distance for each nearest neighbor pair.

[0096] Furthermore, let the segmentation threshold be the first... The nearest neighbor distance is used to classify non-ground point cloud data pairs within the nearest neighbor data pair whose nearest neighbor distance is greater than the segmentation threshold as low-density data pairs; and non-ground point cloud data pairs within the nearest neighbor data pair whose nearest neighbor distance is less than or equal to the segmentation threshold as high-density data pairs.

[0097] Furthermore, the segmentation threshold is obtained as the first... The segmentation effectiveness parameter for the nearest neighbor distance is calculated using the following formula:

[0098] ;

[0099] In the formula, The segmentation threshold is the first... The segmentation effectiveness parameter when the nearest neighbor distance is given. This indicates the number of data points belonging to high-density data pairs. This indicates the number of data points belonging to low-density data pairs. For the first The number of nearest neighbor data point pairs whose nearest neighbor distance is less than the segmentation threshold within all nearest neighbor data pairs of a data point belonging to a high-density data pair; For the first The number of nearest neighbor pairs whose nearest neighbor distance is greater than the segmentation threshold within all nearest neighbor pairs of a data point belonging to a low-density data pair. Where, if the... If a non-ground point cloud data point belongs to both high-density and low-density data pairs, then calculate... At that time, the first Each non-ground point cloud data point is considered as a data point belonging to a high-density data pair, and calculations are performed. At that time, the first Each non-ground point cloud data point is considered a data point belonging to a low-density data pair.

[0100] It should be noted that, The larger the value, the more likely the segmentation threshold is to be the first. When the nearest neighbor distance is 1, it more closely matches the characteristic that other non-ground point cloud data around high-density data are mostly high-density data, i.e., the 1st nearest neighbor distance. The greater the probability that the nearest neighbor distance is the optimal segmentation threshold, the higher the probability that the nearest neighbor distance is the optimal segmentation threshold. The larger the value, the more it matches the characteristic that the surrounding non-ground point cloud data is mostly low-density data, i.e., the higher the value. The greater the probability that the nearest neighbor distance is the optimal segmentation threshold, the higher the likelihood.

[0101] pass The calculation formula, find the one that makes When the value is at its maximum, it represents the distance between the nearest neighbor pairs. This distance is then used to classify the nearest neighbor data pairs into high-density and low-density pairs.

[0102] Ideally, high-density and low-density segmentation should occur when the nearest neighbors of high-density data points (nearest neighbors being the points with the closest Euclidean distance to a given point, and there may be multiple nearest neighbors) are all high-density data points, and the nearest neighbors of low-density data points are all low-density data points. When this reaches its maximum value, it means that more neighboring point pairs belong to the same density category, thus making the division between high-density and low-density areas more coherent and reasonable.

[0103] Furthermore, the segmentation effectiveness parameter is calculated when the segmentation threshold is the distance between each nearest neighbor, and the nearest neighbor distance corresponding to the maximum segmentation effectiveness is recorded as the optimal segmentation threshold.

[0104] Furthermore, non-ground point cloud data within nearest neighbor pairs with a distance greater than the optimal segmentation threshold are denoted as final low-density data; non-ground point cloud data within nearest neighbor pairs with a distance less than or equal to the optimal segmentation threshold are denoted as final high-density data. The partitioning results of high-density and low-density data pairs are as follows: Figure 10 As shown in the figure, high-density data pairs are represented in blue and connected by yellow straight lines; low-density data pairs are represented in red and connected by black straight lines.

[0105] At this point, all non-ground data is divided into final high-density data and final low-density data.

[0106] Step S003: Using a non-dominated sorting genetic algorithm, obtain DBSCAN clustering algorithms for low-density parameter combinations and high-density parameter combinations within multiple parameter combinations. While ensuring the core points are the final high-density data, use the DBSCAN clustering algorithm for high-density parameter combinations within each parameter combination to cluster all non-ground point cloud data, obtaining noise points and high-density clusters within each parameter combination. While ensuring the core points are the final low-density data, use the DBSCAN clustering algorithm for low-density parameter combinations within each parameter combination to cluster the noise points within that parameter combination, obtaining low-density clusters and secondary noise points within each parameter combination. Based on the distances between the secondary noise points and the low-density clusters, as well as the non-ground point cloud data within the high-density clusters, the final cluster within each parameter combination is obtained.

[0107] It should be noted that when combining the non-dominated sorting genetic algorithm with the DBSCAN clustering algorithm, the non-dominated sorting genetic algorithm is first used to generate multiple parameter combinations of neighborhood radii and minimum number of points within the DBSCAN clustering algorithm. Then, the clustering results for each parameter combination are obtained. Using existing methods for evaluating clustering results, the fitness of each clustering result is obtained. Finally, based on the fitness of the clustering results under each parameter combination, a relatively optimal parameter combination is determined.

[0108] It should be further explained that the DBSCAN clustering algorithm, also known as a density-based clustering algorithm, essentially clusters data based on the number of surrounding data points for each data point. However, when acquiring 3D point cloud data using LiDAR, the amount of point cloud data for each object is related to its distance from the LiDAR. This means that when using the DBSCAN clustering algorithm, some point cloud data points of objects far from the LiDAR will be considered noise points. Therefore, this invention, when clustering non-ground point cloud data using the DBSCAN clustering algorithm, first clusters all non-ground point cloud data based on a set of parameter combinations and the minimum number of points. Then, the DBSCAN clustering algorithm is used again to cluster the noise points from the previous clustering result. Based on the two clustering results, the final clustering result is obtained.

[0109] It's important to further clarify that the secondary clustering of noisy points within the initial clustering results is necessary because the amount of 3D point cloud data from the objects collected by LiDAR is inversely proportional to the distance between the LiDAR and the LiDAR. This means that after the initial clustering, some non-ground objects located far from the LiDAR might be considered noise points. Therefore, when re-clustering noisy points from the initial clustering results, the core points must be low-density data; conversely, when performing the initial clustering on all non-ground point cloud data, the core points must be high-density data. When using the DBSCAN clustering algorithm, to avoid incorrectly grouping two clusters into one, the minimum neighborhood radius must be controlled. However, this prevents sparser points from being successfully clustered, causing some sparse points to be identified as noise points. Figure 7 This is a schematic diagram showing the distribution of noise points and data points within clusters after clustering, further as follows: Figure 7 In Figure (a), the green dots represent one cluster, the blue dots represent another, and the black dots represent noise points. However, these black dots may not actually be noise points, but rather formed due to an unreasonable neighborhood radius setting. Therefore, based on the distance between each noise point and the data points in other surrounding clusters, the noise points are redistributed, allowing data points mistakenly identified as noise points to be returned to their respective clusters without altering the neighborhood radius. The image after redistributing the noise points is shown below. Figure 7 The clustering result image after redistributing the noise points in figure (a) is as follows: Figure 7 Figure (b).

[0110] Specifically, a non-dominated sorting genetic algorithm is used to obtain multiple parameter combinations of neighborhood radius and minimum number of points. Each parameter combination includes a high-density combination of neighborhood radius and minimum number of points, and a low-density combination of neighborhood radius and minimum number of points.

[0111] Furthermore, the neighborhood radius and minimum number of points under the high-density data parameter combination within the y-th parameter combination are taken as the neighborhood radius and minimum number of points within the DBSCAN clustering algorithm, thus obtaining the DBSCAN clustering algorithm under the high-density parameter combination within the y-th parameter combination.

[0112] Furthermore, the neighborhood radius and minimum number of points under the low-density data parameter combination within the y-th parameter combination are taken as the neighborhood radius and minimum number of points within the DBSCAN clustering algorithm, thus obtaining the DBSCAN clustering algorithm under the low-density parameter combination within the y-th parameter combination.

[0113] Furthermore, while ensuring that the core points are the final high-density data, the DBSCAN clustering algorithm under the high-density parameter combination within the y-th parameter combination is used as the distance metric to cluster all non-ground point cloud data, resulting in the cluster, core points, and noise points under the high-density parameter combination within the y-th parameter combination. These are denoted as the high-density cluster within the y-th parameter combination, and the noise points and core points within this parameter combination.

[0114] Furthermore, while ensuring that the core points are the final low-density data, the DBSCAN clustering algorithm under the low-density parameter combination within the y-th parameter combination is used to continue clustering the noise points within the y-th parameter combination. The clusters, noise points, and core points obtained from this clustering are respectively denoted as the low-density clusters within the y-th parameter combination, the secondary noise points within the y-th parameter combination, and the core points within the y-th parameter combination.

[0115] Furthermore, the y-th parameter combination is... The Euclidean distance between each secondary noise point and each data point within the initial cluster and the noise cluster is denoted as the distance between the y-th parameter combination and the y-th data point. The difference distance between the secondary noise points.

[0116] Furthermore, if the y-th parameter combination contains the th... If the minimum difference distance of the _th secondary noise points is less than or equal to the neighborhood radius of the high-density data parameter combination within the _th parameter combination, then the _th parameter combination within the _th parameter combination will be... Each secondary noise point is placed within the cluster of data corresponding to its minimum difference distance.

[0117] Furthermore, the above operation is performed on all secondary noise points within the y-th parameter combination to obtain the updated low-density cluster and high-density cluster within the y-th parameter combination, and these are recorded as the final cluster within the y-th parameter combination.

[0118] Step S004: Based on the distribution of data around the core points in the final cluster within each parameter combination and the distribution of core points in different clusters, obtain the first reference value and the second reference value of the final clustering result within each parameter combination.

[0119] It should be noted that currently, when evaluating clustering results, the distance from data within a cluster to the centroid of that cluster, as well as the distance between centroids of different clusters, are used as parameters for evaluating the clustering results. However, this method is not suitable for judging the clustering results of density-based clustering algorithms. Figure 5 This is a diagram illustrating the results of clustering data using the DBSCAN clustering algorithm. For example, the accurate clustering results when using the DBSCAN algorithm are shown below. Figure 5As shown in Figure (a), since the centroids of the two clusters are both located at the center, and the distance between the centroids of the two clusters is very close, it is obvious that traditional evaluation metrics are ineffective in measuring this clustering result. However, as... Figure 5 In the case shown in Figure (b), the centroids of the two clusters are located on opposite sides, resulting in a greater distance between the two clusters. Furthermore, points within a cluster are closer to its centroid, leading traditional evaluation metrics to consider this clustering result to be of high quality. Therefore, existing evaluation methods for clustering results need to be improved. Figure 8 This is a diagram showing the results of clustering non-terrestrial point cloud data using existing methods.

[0120] It should be further explained that when using the DBSCAN clustering algorithm to cluster data, it first determines whether a data point is a core point by checking if the number of data points within its neighborhood radius is greater than the minimum number of points. Then, each core point and its neighborhood radius are considered as a cluster, completing the initial clustering. Subsequently, if a core point is within the cluster of another core point, the clusters containing both core points are considered as the same cluster, completing the initial merging of clusters. In essence, the DBSCAN clustering algorithm treats densely connected data as a single cluster. Therefore, the mean of the Euclidean distance between a core point and all data points within its neighborhood radius is used as the density estimate of that core point. Then, the variance of the density estimates of all core points within a cluster is used to evaluate the density consistency of that cluster. That is, this invention uses the sum of the density consistency of all clusters within the clustering result as an evaluation criterion for the clustering result.

[0121] It should be further noted that when judging the clustering effect by the variance of the density estimates of the core points within all clusters in each cluster, it is found that the variance of the density estimates of the core point within each cluster is minimized when there is only one core point. In other words, judging the clustering effect by the density variance of the core points within a cluster may result in over-clustering. Figure 6 This is a diagram illustrating over-clustering. Figure 6 Each data point of different color represents a cluster, meaning that data that should have been in one cluster was incorrectly divided into multiple smaller clusters. These clusters are then separated by noise points. Therefore, by calculating the distance between the core points within different clusters during each clustering process, the performance of two clusters belonging to the same cluster is subdivided and scored. Then, based on the distribution of noise points, a second reference value within the fitness range of the clustering results is obtained.

[0122] It should be further explained that when clustering data using the DBSCAN clustering algorithm, the core points are first determined based on whether the number of data points within the neighborhood radius of each data point is greater than the minimum number of points. Then, each core point and the data within its neighborhood radius are considered as a cluster. That is, during the first clustering stage using the DBSCAN clustering algorithm, all data within the neighborhood radius of each core point are considered as belonging to the same cluster. Considering the characteristic that a data point cannot belong to two clusters simultaneously, this invention considers two clusters to be adjacent if the Euclidean distance between the core points of two different clusters is less than or equal to twice their neighborhood radius. The difference between the Euclidean distance between the core points of these two different clusters and the neighborhood radius under twice the high-density parameter combination is then used as a parameter for calculating the subdivision score. Furthermore, the DBSCAN clustering algorithm clusters the data based on the distribution of the data around each data point. In other words, the closer the density variances of two clusters are, the greater the probability that the two clusters belong to the same cluster. Therefore, the difference in density variances between two clusters is used as a parameter for calculating the over-segmentation score.

[0123] It should be further noted that in real-world clustering, the likelihood of a large number of noisy points within a dataset is relatively low. Therefore, if there are too many noisy points in the clustering result, a larger weight is assigned to the over-subdivision score; conversely, if there are too few noisy points, a larger weight is assigned to the proportion of noisy points, and a smaller weight is assigned to the over-subdivision score, thus obtaining the second parameter for evaluating the effectiveness of the clustering result.

[0124] Specifically, it will be combined with the y-th parameter within the first... Within the final cluster, the first The data whose Euclidean distance to each core point is less than or equal to the neighborhood radius under the high-density parameter combination within the y-th parameter combination is denoted as the data within the y-th parameter combination. Within the final cluster, the first The neighboring points of the core point.

[0125] Combine the y-th parameter into the th... Within the final cluster, the first The mean of the Euclidean distances between each core point and all its neighboring points is used as the y-th parameter combination. Within the final cluster, the first Density estimates for each core point.

[0126] Combine the y-th parameter into the th... The variance of the density estimates of all core points within the final cluster is denoted as the variance of the nth parameter combination within the yth parameter combination. Density consistency of the final clusters.

[0127] The sum of the density consistency of all final clusters within the y-th parameter combination is denoted as the first reference value of the final clustering result within the y-th parameter combination.

[0128] Furthermore, calculate the first parameter combination within the y-th parameter combination. Each core point within the final cluster is related to the first... The distance to each core point within the final cluster. If the y-th parameter combination... Within the final cluster, the first Within the combination of the _ core point and the y-th parameter, the _ _ Within the final cluster, the first If the Euclidean distance between the core points of two clusters is less than 2 times the neighborhood radius under the high-density parameter combination within the y-th parameter combination, then the two clusters are considered adjacent. The difference between the Euclidean distance between the core points of these two different clusters and the neighborhood radius under the high-density parameter combination (2 times the y-th parameter combination) is then used as a parameter for over-subdivision scoring. Furthermore, the DBSCAN clustering algorithm essentially clusters data based on the distribution of data surrounding each data point. That is, the closer the density variances of two clusters are, the greater the probability that they belong to the same cluster. Therefore, the difference in density variances between two clusters is used as a parameter for calculating the over-subdivision score.

[0129] Furthermore, obtain the first parameter within the y-th parameter combination. The final cluster and the first The specific formula for calculating the subdivision score of the final cluster is as follows:

[0130] ;

[0131] In the formula, Within the y-th parameter combination, the th... The final cluster and the first The subdivision score of the final cluster, Indicates the y-th parameter combination. The variance of the density estimates of the core points within each final cluster Indicates the y-th parameter combination. The variance of the density estimate of the core within each final cluster. Indicates the y-th parameter combination. The final cluster and the first The number of subdivisions in the final cluster. This represents the neighborhood radius of the high-density data parameter combination within the y-th parameter combination. Indicates the y-th parameter combination. The final cluster and the first The first final cluster Each level of detailed performance This represents the absolute value function.

[0132] In the formula, Indicates the y-th parameter combination. The final cluster and the first The difference in intra-cluster density variance between the final clusters is used to measure the... The final cluster and the first Density-consistent similarity among the final clusters.

[0133] This is the simplified result; the original expression is... The numerator represents the difference between the Euclidean distance between the core point a in the f-th final cluster and the core point b in the d-th final cluster, and the neighborhood radius under the high-density data parameter combination multiplied by 2, because this difference is at most 2. The minimum value is 0, so the purpose of the denominator here is to normalize. The final simplified form is... This involves calculating the similarity in distance between two core points.

[0134] The degree of subdivision of the two clusters is measured by the distance between the core points. If there are multiple core points in the two clusters whose distance is less than twice the neighborhood radius, then it is more likely that the two clusters are actually a single cluster. This represents the difference in density consistency between the two clusters. In other words, the smaller the difference in density consistency between cluster A and cluster B, the higher the similarity in intra-cluster density between cluster A and cluster B.

[0135] In general, it is through We measure the similarity of density uniformity and the proximity of core points within any two clusters to determine if they should belong to the same cluster but have been over-segmented. The larger the value, the more severe the over-segmentation.

[0136] Input the sum of the over-subdivision scores of all two different final clusters within the y-th parameter combination into the... Inside the function, the output is recorded as the over-subdivision score of the final clustering result within the y-th parameter combination.

[0137] The following is a further explanation of the overly detailed scoring:

[0138] Because some parameter combinations can result in clustering that divides data that should belong to a single cluster into several smaller, denser clusters, this invention proposes an over-segmentation score (as part of a second reference value) to avoid this situation. The over-segmentation score measures the degree to which data that should belong to a single cluster is incorrectly clustered into different clusters.

[0139] Depend on Figure 5 (b) and Figure 6 It can be seen that these should have been clustered into one cluster, but due to the incorrect selection of parameter combinations, the following result was obtained: Figure 6 The clustering results shown are due to the DBSCAN algorithm clustering all density-connected data, but it is clear that... Figure 6 These small clusters are formed because of over-segmentation, resulting in multiple small and dense clusters. In other words, without incorporating the over-segmentation score as a reference value, the NSGAII algorithm's automatic parameter acquisition could easily lead to clustering results such as... Figure 6 The parameter combinations shown.

[0140] Adding the over-segmentation score as a reference value can effectively prevent data that was originally in one cluster from being incorrectly clustered into multiple clusters. This situation is called over-segmentation. The smaller the over-segmentation score, the less severe the over-segmentation.

[0141] Furthermore, the formula for calculating the noise ratio of the final clustering result within the y-th parameter combination is as follows:

[0142] ;

[0143] In the formula, This represents the noise proportion of the final clustering result within the y-th parameter combination. This indicates the amount of non-terrestrial point cloud data. This represents the number of non-ground point cloud data that are not in the final cluster within the y-th parameter combination.

[0144] Furthermore, the specific calculation formula for obtaining the second reference value within the fitness of the clustering results under the high-density parameter combination within the y-th parameter combination is as follows:

[0145] ;

[0146] In the formula, The second parameter represents the final clustering result within the y-th parameter combination. This represents the over-subdivision score of the final clustering result within the y-th parameter combination. This represents the noise proportion in the final clustering result within the y-th parameter combination. "Indicates the proportion of non-noise data points, This indicates the proportion of noise; add it here. This is to control the proportion of noise. If the proportion of noise is too high, then... The value will also increase accordingly, which can effectively avoid a combination of parameters that results in a large proportion of noise in the clustering results.

[0147] Step S005: Input the first and second reference values ​​of the final clustering results within each parameter combination as two fitness values ​​into the non-dominated sorting genetic algorithm to obtain the optimal parameter combination when using the DBSCAN clustering algorithm to cluster non-ground point cloud data; perform multiple clustering operations on all non-ground point cloud data using the DBSCAN clustering algorithm with the optimal parameter combination when using the DBSCAN clustering algorithm to cluster non-ground point cloud data to obtain the final clustering result, thus completing the segmentation of the non-ground point cloud data.

[0148] Specifically, the first and second reference values ​​of the final clustering results within each high-density parameter combination are used as the two fitness values ​​of that parameter combination and input into the non-dominated sorting genetic algorithm to obtain the optimal parameter combination when using the DBSCAN clustering algorithm to cluster non-terrestrial point cloud data.

[0149] The neighborhood radius and minimum number of points in the optimal parameter combination when using the DBSCAN clustering algorithm to cluster non-ground point cloud data are used as the neighborhood radius and minimum number of points in the DBSCAN clustering algorithm, thus obtaining the DBSCAN clustering algorithm under the high-density parameter combination within the optimal parameter combination.

[0150] While ensuring that the core points are the final high-density data, the DBSCAN clustering algorithm under the high-density parameter combination within the optimal parameter combination is used as the distance metric to cluster all non-ground point cloud data, and the clusters and noise points under the high-density parameter combination within the optimal parameter combination are obtained, which are denoted as the initial clusters and initial noise points, respectively.

[0151] The neighborhood radius and minimum number of points under the low-density data parameter combination within the optimal parameter combination when clustering all non-ground point cloud data are taken as the neighborhood radius and minimum number of points within the DBSCAN clustering algorithm, thus obtaining the DBSCAN clustering algorithm under the low-density parameter combination within the optimal parameter combination.

[0152] While ensuring that the core points are the final low-density data, the DBSCAN clustering algorithm under the low-density parameter combination within the optimal parameter combination is used as the distance metric to cluster all non-ground point cloud data, and the clusters and noise points under the low-density parameter combination within the optimal parameter combination are obtained, which are denoted as noise clusters and secondary noise, respectively.

[0153] Furthermore, the first The Euclidean distance between the nth secondary noise point and each data point within the initial cluster and the noise cluster is denoted as the nth noise point. The difference distance between the two secondary noise points. If the first If the minimum difference distance of the nth secondary noise points is less than or equal to the best neighborhood when clustering the initial noise points, then the nth... Each secondary noise point is placed in the cluster corresponding to the data with the minimum difference distance. This process is repeated for all secondary noise points to obtain updated initial clusters and updated noise clusters. Based on these updated initial and noise clusters, the final clustering result is obtained. The final clustering result is used to complete the segmentation of the non-terrestrial point cloud data. Figure 9 This is a diagram showing the result of clustering non-terrestrial point cloud data using the method of the present invention.

[0154] This invention employs a secondary clustering and distance verification approach to avoid misclassifying low-density objects as noise, reducing missed classifications. Simultaneously, it eliminates true noise, lowering the misclassification rate and making the segmentation results of non-terrestrial point clouds closer to the distribution of objects in the actual scene (e.g., nearby vehicles, distant signs, and other objects of varying densities can be correctly classified). By first partitioning, then performing targeted clustering, and finally verifying and optimizing, this process effectively overcomes the impact of uneven point cloud density on DBSCAN clustering, improving the accuracy of non-terrestrial point cloud segmentation.

[0155] Another embodiment of the present invention provides a point cloud data segmentation system based on the DBSCAN clustering algorithm, comprising:

[0156] The point cloud data acquisition module is used to acquire non-ground point cloud data.

[0157] The point cloud data segmentation module is used to divide non-ground point cloud data into final high-density data and final low-density data based on the distribution of non-ground point cloud data.

[0158] While ensuring that the core points are the final high-density data, the DBSCAN clustering algorithm is used to cluster all non-ground point cloud data to obtain initial clusters and initial noise points.

[0159] While ensuring that the core points are the final low-density data, the DBSCAN clustering algorithm is used to cluster the initial noise points to obtain noise clusters and secondary noise points.

[0160] The final clustering result is obtained based on the Euclidean distance between the secondary noise points and the initial cluster, as well as the distance between each data point within the noise cluster, thus completing the segmentation of the non-terrestrial point cloud data.

[0161] It should be noted that the specific embodiments described above enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. Therefore, although the present invention has been described in detail in this specification and embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention; and all technical solutions and improvements that do not depart from the spirit and scope of the present invention are covered within the protection scope of the present invention patent.

Claims

1. A point cloud data segmentation method based on the DBSCAN clustering algorithm, characterized in that, include: Acquire non-terrestrial point cloud data; Based on the distribution of non-ground point cloud data, the non-ground point cloud data is divided into final high-density data and final low-density data. While ensuring that the core points are the final high-density data, the DBSCAN clustering algorithm is used to cluster all non-ground point cloud data to obtain initial clusters and initial noise points. While ensuring that the core points are the final low-density data, the DBSCAN clustering algorithm is used to cluster the initial noise points to obtain noise clusters and secondary noise points; Based on the Euclidean distance between the secondary noise points and the initial cluster, as well as the distance between each data point within the noise cluster, the final clustering result is obtained, thus completing the segmentation of the non-ground point cloud data. It also includes combining the non-dominated sorting genetic algorithm with the DBSCAN clustering algorithm to obtain the optimal parameter combination when using the DBSCAN clustering algorithm to cluster non-ground point cloud data, and obtaining the final segmentation result by using the optimal parameter combination when using the DBSCAN clustering algorithm to cluster non-ground point cloud data. The optimal parameter combination used in the DBSCAN clustering algorithm to cluster non-ground point cloud data to obtain the final segmentation result is as follows: Using a non-dominated sorting genetic algorithm, multiple parameter combinations are obtained. Each parameter combination contains a high-density data parameter combination with a neighborhood radius and a minimum number of points, and a low-density parameter combination with a neighborhood radius and a minimum number of points. The neighborhood radius and minimum number of points in the high-density parameter combination are different from those in the low-density parameter combination. The neighborhood radius and minimum number of points under the high-density data parameter combination within the y-th parameter combination are taken as the neighborhood radius and minimum number of points within the DBSCAN clustering algorithm, thus obtaining the DBSCAN clustering algorithm under the high-density parameter combination within the y-th parameter combination. The neighborhood radius and minimum number of points under the low-density data parameter combination within the y-th parameter combination are taken as the neighborhood radius and minimum number of points within the DBSCAN clustering algorithm, thus obtaining the DBSCAN clustering algorithm under the low-density parameter combination within the y-th parameter combination. While ensuring that the core points are the final high-density data, the DBSCAN clustering algorithm under the high-density parameter combination within the y-th parameter combination is used as the distance metric to cluster all non-ground point cloud data, and the clusters, core points and noise points under the high-density parameter combination within the y-th parameter combination are obtained. These are denoted as the high-density cluster within the y-th parameter combination, and the noise points and core points within this parameter combination. While ensuring that the core points are the final low-density data, the DBSCAN clustering algorithm under the low-density parameter combination in the y-th parameter combination is used to continue clustering the noise points in the y-th parameter combination. The clusters, noise points and core points obtained from this clustering are respectively denoted as the low-density clusters in the y-th parameter combination, the secondary noise points in the y-th parameter combination and the core points in the y-th parameter combination. Combine the y-th parameter into the th... The Euclidean distance between each secondary noise point and each data point within the initial cluster and the noise cluster is denoted as the distance between the y-th parameter combination and the y-th data point. The difference distance between each secondary noise point; If the y-th parameter combination contains the th... If the minimum difference distance of the _th secondary noise points is less than or equal to the neighborhood radius of the high-density data parameter combination within the _th parameter combination, then the _th parameter combination within the _th parameter combination will be... Each secondary noise point is placed within the cluster of data whose minimum difference distance corresponds to its own. Perform the above operation on all secondary noise points in the y-th parameter combination to obtain the updated low-density cluster and high-density cluster in the y-th parameter combination, and record them as the final cluster in the y-th parameter combination. By analyzing the distribution of data around the core points within the final cluster within the y-th parameter combination, we obtain the first reference value of the final clustering result within the y-th parameter combination. Based on the distance between the core points in different final clusters within the y-th parameter combination, the distribution differences of the data around the core points, and the distribution of noise points, a second reference value for the final clustering result within the y-th parameter combination is obtained. The first and second reference values ​​of the final clustering result in each parameter combination are used as the two fitness values ​​of that parameter combination and input into the non-dominated sorting genetic algorithm to obtain the optimal parameter combination when using the DBSCAN clustering algorithm to cluster non-ground point cloud data. When using the DBSCAN clustering algorithm to cluster non-ground point cloud data, the neighborhood radius and minimum number of points in the optimal parameter combination are used as the neighborhood radius and minimum number of points in the DBSCAN clustering algorithm, thus obtaining the DBSCAN clustering algorithm under the high-density parameter combination within the optimal parameter combination. While ensuring that the core points are the final high-density data, the DBSCAN clustering algorithm under the high-density parameter combination within the optimal parameter combination is used as the distance metric to cluster all non-ground point cloud data, and the clusters and noise points under the high-density parameter combination within the optimal parameter combination are obtained, which are denoted as the initial clusters and initial noise points, respectively. The neighborhood radius and minimum number of points under the low-density data parameter combination within the optimal parameter combination when clustering all non-ground point cloud data are taken as the neighborhood radius and minimum number of points within the DBSCAN clustering algorithm, thus obtaining the DBSCAN clustering algorithm under the low-density parameter combination within the optimal parameter combination. While ensuring that the core points are the final low-density data, the DBSCAN clustering algorithm under the low-density parameter combination within the optimal parameter combination is used as the distance metric to cluster all non-ground point cloud data, and the clusters and noise points under the low-density parameter combination within the optimal parameter combination are obtained, which are denoted as noise clusters and secondary noise points, respectively. The final clustering result is obtained based on the Euclidean distance between the secondary noise points and the initial cluster, as well as the distance between each data point within the noise cluster.

2. The point cloud data segmentation method based on the DBSCAN clustering algorithm according to claim 1, characterized in that, The process of dividing the non-ground point cloud data into final high-density data and final low-density data based on the distribution of the non-ground point cloud data is as follows: Calculate the first The Euclidean distance between the nth non-ground point cloud data and all non-ground point cloud data is obtained. A distance set of non-terrestrial point cloud data; The first Let denoted as the set of all data corresponding to the minimum distance within the distance set of the nth non-ground point cloud data. The nearest neighbor data of a non-terrestrial point cloud data; Through the first The non-ground point cloud data and its first Given a set of nearest neighbor data, obtain a nearest neighbor pair. The Euclidean distance between the two data points in each nearest neighbor pair is recorded as the nearest neighbor distance of that nearest neighbor pair, thus obtaining the nearest neighbor distance set; Each nearest neighbor distance in the nearest neighbor distance set is used as a segmentation threshold, and high-density data and low-density data under each segmentation threshold are obtained based on the segmentation threshold. Based on the high-density and low-density data within each nearest neighbor pair under each segmentation threshold, the final high-density data and the final low-density data are obtained.

3. The point cloud data segmentation method based on the DBSCAN clustering algorithm according to claim 2, characterized in that, The specific steps for obtaining the high-density data and low-density data under each segmentation threshold are as follows: Let the segmentation threshold be the first... The nearest neighbor distance is defined as the non-ground point cloud data pair within the nearest neighbor data pair whose nearest neighbor distance is greater than the segmentation threshold. Non-ground point cloud data pairs within the nearest neighbor data pairs whose nearest neighbor distance is less than or equal to the segmentation threshold are denoted as high-density data pairs.

4. The point cloud data segmentation method based on the DBSCAN clustering algorithm according to claim 2, characterized in that, The specific steps for obtaining the final high-density data and the final low-density data based on the high-density data and low-density data within each nearest neighbor data pair under each segmentation threshold are as follows: Obtain the segmentation threshold as the first The segmentation effectiveness parameter for the nearest neighbor distance is calculated using the following formula: ； In the formula, The segmentation threshold is the first... The segmentation effectiveness parameter when the nearest neighbor distance is given. This indicates the number of data points belonging to high-density data pairs. This indicates the number of data points belonging to low-density data pairs. For the first The number of nearest neighbor data point pairs whose nearest neighbor distance is less than the segmentation threshold within all nearest neighbor data pairs of a data point belonging to a high-density data pair; For the first The number of nearest neighbor pairs whose nearest neighbor distance is greater than the segmentation threshold within all nearest neighbor pairs of data points belonging to low-density data pairs; The nearest neighbor distance corresponding to the maximum segmentation effectiveness is denoted as the optimal segmentation threshold; Non-ground point cloud data within the nearest neighbor data pair whose nearest neighbor distance is greater than the optimal segmentation threshold are denoted as final low-density data; non-ground point cloud data within the nearest neighbor data pair whose nearest neighbor distance is less than or equal to the optimal segmentation threshold are denoted as final high-density data.

5. The point cloud data segmentation method based on the DBSCAN clustering algorithm according to claim 1, characterized in that, The first reference value of the final clustering result within the y-th parameter combination is obtained by analyzing the distribution of data around the core point within the final cluster within the y-th parameter combination. The specific steps are as follows: The combination with the y-th parameter is the first Within the final cluster, the first The data whose Euclidean distance to each core point is less than or equal to the neighborhood radius under the high-density parameter combination within the y-th parameter combination is denoted as the data within the y-th parameter combination. Within the final cluster, the first Neighborhood points of a core point; Combine the y-th parameter into the th... Within the final cluster, the first The mean of the Euclidean distances between each core point and all its neighboring points is used as the y-th parameter combination. Within the final cluster, the first Density estimates for each core point; Combine the y-th parameter into the th... The variance of the density estimates of all core points within the final cluster is denoted as the variance of the nth parameter combination within the yth parameter combination. The density consistency of each final cluster; The sum of the density consistency of all final clusters within the y-th parameter combination is denoted as the first reference value of the final clustering result within the y-th parameter combination.

6. The point cloud data segmentation method based on the DBSCAN clustering algorithm according to claim 5, characterized in that, The second reference value for the final clustering result within the y-th parameter combination is obtained based on the distance between the core points in different final clusters, the distribution differences of the data around the core points, and the distribution of noise points. The specific steps are as follows: Calculate the y-th parameter combination. Each core point within the final cluster is related to the first... The distance between each core point within each final cluster; If the y-th parameter combination contains the th... Within the final cluster, the first Within the combination of the _ core point and the y-th parameter, the _ _ Within the final cluster, the first If the Euclidean distance between two core points is less than 2 times the neighborhood radius under the high-density parameter combination within the y-th parameter combination, then the two clusters are considered adjacent. Then, the difference between the Euclidean distance between the core points of these two different clusters and the neighborhood radius under the high-density parameter combination under the y-th parameter combination is used as a parameter for over-subdivision scoring. Get the y-th parameter combination. The final cluster and the first The specific formula for calculating the subdivision score of the final cluster is as follows: ； In the formula, Within the y-th parameter combination, the th... The final cluster and the first The subdivision score of the final cluster, Indicates the y-th parameter combination. The variance of the density estimates of the core points within each final cluster Indicates the y-th parameter combination. The variance of the density estimate of the core within each final cluster. Indicates the y-th parameter combination. The final cluster and the first The number of subdivisions in the final cluster. This represents the neighborhood radius of the high-density data parameter combination within the y-th parameter combination. Indicates the y-th parameter combination. The final cluster and the first The first final cluster Each level of detailed performance Represents the absolute value function; The sum of the over-subdivision scores of all two different final clusters within the y-th parameter combination is input into the normalization function, and the output is recorded as the over-subdivision score of the final clustering result within the y-th parameter combination. The formula for calculating the noise ratio of the final clustering result within the y-th parameter combination is as follows: ； In the formula, This represents the noise proportion of the final clustering result within the y-th parameter combination. This indicates the amount of non-terrestrial point cloud data. This represents the number of non-ground point cloud data points that are not in the final cluster within the y-th parameter combination; The specific calculation formula for obtaining the second reference value within the fitness of the clustering results under the high-density parameter combination within the y-th parameter combination is as follows: ； In the formula, The second parameter represents the final clustering result within the y-th parameter combination. This represents the over-subdivision score of the final clustering result within the y-th parameter combination. This represents the noise proportion of the final clustering result within the y-th parameter combination.

7. The point cloud data segmentation method based on the DBSCAN clustering algorithm according to claim 1, characterized in that, The final clustering result is obtained based on the Euclidean distance between the secondary noise points and the initial cluster, as well as the distance between each data point within the noise cluster. The specific steps are as follows: The first The Euclidean distance between the nth secondary noise point and each data point within the initial cluster and the noise cluster is denoted as the nth noise point. The difference distance between each secondary noise point; If the first If the minimum difference distance of the nth secondary noise points is less than or equal to the neighborhood radius of the DBSCAN clustering algorithm under the optimal high-density parameter combination when clustering all non-ground point cloud data, then the nth... Each secondary noise point is placed within the cluster of data whose minimum difference distance corresponds to its own. Perform the above operation on all secondary noise points to obtain the updated initial cluster and the updated noise cluster; based on the updated initial cluster and the noise cluster, obtain the final clustering result.

8. A point cloud data segmentation system based on the DBSCAN clustering algorithm, characterized in that, include: Point cloud data acquisition module, used to acquire non-ground point cloud data; The point cloud data segmentation module is used to divide the non-ground point cloud data into final high-density data and final low-density data according to the distribution of the non-ground point cloud data. While ensuring that the core points are the final high-density data, the DBSCAN clustering algorithm is used to cluster all non-ground point cloud data to obtain initial clusters and initial noise points. While ensuring that the core points are the final low-density data, the DBSCAN clustering algorithm is used to cluster the initial noise points to obtain noise clusters and secondary noise points; Based on the Euclidean distance between the secondary noise points and the initial cluster, as well as the distance between each data point within the noise cluster, the final clustering result is obtained, thus completing the segmentation of the non-ground point cloud data. It also includes combining the non-dominated sorting genetic algorithm with the DBSCAN clustering algorithm to obtain the optimal parameter combination when using the DBSCAN clustering algorithm to cluster non-ground point cloud data, and obtaining the final segmentation result by using the optimal parameter combination when using the DBSCAN clustering algorithm to cluster non-ground point cloud data. The optimal parameter combination used in the DBSCAN clustering algorithm to cluster non-ground point cloud data to obtain the final segmentation result is as follows: Using a non-dominated sorting genetic algorithm, multiple parameter combinations are obtained. Each parameter combination contains a high-density data parameter combination with a neighborhood radius and a minimum number of points, and a low-density parameter combination with a neighborhood radius and a minimum number of points. The neighborhood radius and minimum number of points in the high-density parameter combination are different from those in the low-density parameter combination. The neighborhood radius and minimum number of points under the high-density data parameter combination within the y-th parameter combination are taken as the neighborhood radius and minimum number of points within the DBSCAN clustering algorithm, thus obtaining the DBSCAN clustering algorithm under the high-density parameter combination within the y-th parameter combination. The neighborhood radius and minimum number of points under the low-density data parameter combination within the y-th parameter combination are taken as the neighborhood radius and minimum number of points within the DBSCAN clustering algorithm, thus obtaining the DBSCAN clustering algorithm under the low-density parameter combination within the y-th parameter combination. While ensuring that the core points are the final high-density data, the DBSCAN clustering algorithm under the high-density parameter combination within the y-th parameter combination is used as the distance metric to cluster all non-ground point cloud data, and the clusters, core points and noise points under the high-density parameter combination within the y-th parameter combination are obtained. These are denoted as the high-density cluster within the y-th parameter combination, and the noise points and core points within this parameter combination. While ensuring that the core points are the final low-density data, the DBSCAN clustering algorithm under the low-density parameter combination in the y-th parameter combination is used to continue clustering the noise points in the y-th parameter combination. The clusters, noise points and core points obtained from this clustering are respectively denoted as the low-density clusters in the y-th parameter combination, the secondary noise points in the y-th parameter combination and the core points in the y-th parameter combination. Combine the y-th parameter into the th... The Euclidean distance between each secondary noise point and each data point within the initial cluster and the noise cluster is denoted as the distance between the y-th parameter combination and the y-th data point. The difference distance between each secondary noise point; If the y-th parameter combination contains the th... If the minimum difference distance of the _th secondary noise points is less than or equal to the neighborhood radius of the high-density data parameter combination within the _th parameter combination, then the _th parameter combination within the _th parameter combination will be... Each secondary noise point is placed within the cluster of data whose minimum difference distance corresponds to its own. Perform the above operation on all secondary noise points in the y-th parameter combination to obtain the updated low-density cluster and high-density cluster in the y-th parameter combination, and record them as the final cluster in the y-th parameter combination. By analyzing the distribution of data around the core points within the final cluster within the y-th parameter combination, we obtain the first reference value of the final clustering result within the y-th parameter combination. Based on the distance between the core points in different final clusters within the y-th parameter combination, the distribution differences of the data around the core points, and the distribution of noise points, a second reference value for the final clustering result within the y-th parameter combination is obtained. The first and second reference values ​​of the final clustering result in each parameter combination are used as the two fitness values ​​of that parameter combination and input into the non-dominated sorting genetic algorithm to obtain the optimal parameter combination when using the DBSCAN clustering algorithm to cluster non-ground point cloud data. When using the DBSCAN clustering algorithm to cluster non-ground point cloud data, the neighborhood radius and minimum number of points in the optimal parameter combination are used as the neighborhood radius and minimum number of points in the DBSCAN clustering algorithm, thus obtaining the DBSCAN clustering algorithm under the high-density parameter combination within the optimal parameter combination. While ensuring that the core points are the final high-density data, the DBSCAN clustering algorithm under the high-density parameter combination within the optimal parameter combination is used as the distance metric to cluster all non-ground point cloud data, and the clusters and noise points under the high-density parameter combination within the optimal parameter combination are obtained, which are denoted as the initial clusters and initial noise points, respectively. The neighborhood radius and minimum number of points under the low-density data parameter combination within the optimal parameter combination when clustering all non-ground point cloud data are taken as the neighborhood radius and minimum number of points within the DBSCAN clustering algorithm, thus obtaining the DBSCAN clustering algorithm under the low-density parameter combination within the optimal parameter combination. While ensuring that the core points are the final low-density data, the DBSCAN clustering algorithm under the low-density parameter combination within the optimal parameter combination is used as the distance metric to cluster all non-ground point cloud data, and the clusters and noise points under the low-density parameter combination within the optimal parameter combination are obtained, which are denoted as noise clusters and secondary noise points, respectively. The final clustering result is obtained based on the Euclidean distance between the secondary noise points and the initial cluster, as well as the distance between each data point within the noise cluster.

Images